Improved Steel Surface Defect Detection Algorithm Based on YOLOv8

Abstract

An enhanced steel surface defect detection algorithm based on YOLOv8 was introduced to enhance the accuracy of small target detection. This algorithm incorporates an attention-free mechanism to calculate attention-weight, aiding in the extraction of specific feature regions. Additionally, improvements were made to the SPPF module to expand the receptive field and enhance target detection optimization. Experimental evaluations on the NEU-DET dataset demonstrated significant enhancements over the original YOLOv8 algorithm. The improved algorithm exhibited a 9.3 percentage point increase in precision, a 10 percentage point increase in recall, a 4.6 percentage point increase in [email protected], and a remarkable 21.2 percentage point increase in [email protected]:0.95.Significant progress has also been made in analyzing the surface data of aluminum sheets. The enhanced algorithm has shown a 6% increase in precision compared to the original YOLOv8 algorithm. Additionally, recall has improved by 3.2%, [email protected] has increased by 4.1%, and [email protected]:0.95 has seen a notable rise of 17.4%.